Learning to regulate 3D head shape by removing occluding hair from in-the-wild images

Getting Started

Cloning the repository

https://github.com/vanquish630/BaldMM.git
cd BaldMM

Requirements

1. Python 3.7 (numpy, skimage, scipy, opencv)
2. PyTorch >= 1.8.0
3. Pytorch3d, installation instructions here

Alternatively you can run

pip install -r requirements.txt

How to use

Downloading pretrained models

Download the following trained models and place them in the ./data folder.

1. Trained 3DMM model
2. Trained custom landmark model

Downloading the FLAME texture model

1. Download the FLAME texture space from the FLAME texture space and place it in ./data folder.

2. Modify the flame_tex_path parameter in the ./lib/utils/config file with the path of the downloaded texture space.

Run the demo

 python demos/demo.py -i input -s results --saveObj True --useTex True --device gpu

Where i is the input folder path and s the savefolder path. Please run python demos/demo_reconstruct.py --help for more details.

Acknowledgements

We have used code from the following repositories:

• DECA
• FLAME_PyTorch and TF_FLAME for the FLAME model
• Pytorch3D, neural_renderer, SoftRas for rendering
• kornia for diffrentiable image processing
• face_segmentation for skin mask
• VGGFace2-pytorch for identity loss